Update app.py

app.py

@@ -3,7 +3,6 @@ import streamlit as st
 from transformers import pipeline
 import os
 import tempfile
-import numpy as np
 
 # function part
 # img2text
@@ -43,17 +42,11 @@ def text2story(text):
     
     return story_text
 
-# text2audio - REVISED to correctly handle the audio output
+# text2audio - REVISED to handle audio format correctly
 def text2audio(story_text):
     try:
-        # Use a different TTS model that works reliably with pipeline
-        synthesizer = pipeline("text-to-speech", model="microsoft/speecht5_tts")
-        
-        # Additional input required for this model
-        speaker_embeddings = pipeline(
-            "audio-classification", 
-            model="microsoft/speecht5_speaker_embeddings"
-        )("some_audio_file.mp3")["logits"]
+        # Use a simple, reliable TTS model
+        synthesizer = pipeline("text-to-speech", model="facebook/mms-tts-eng")
         
         # Limit text length to avoid timeouts
         max_chars = 500
@@ -64,98 +57,24 @@ def text2audio(story_text):
             else:
                 story_text = story_text[:max_chars]
         
-        # Generate speech with correct parameters
-        speech = synthesizer(
-            text=story_text,
-            forward_params={"speaker_embeddings": speaker_embeddings}
-        )
+        # Generate speech
+        speech = synthesizer(story_text)
         
-        # Create a temporary WAV file
+        # Create a temporary file with .wav extension
         temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.wav')
         temp_filename = temp_file.name
-        temp_file.close()
-        
-        # Display the structure of the speech output for debugging
-        st.write(f"Speech output keys: {speech.keys()}")
-        
-        # Save the audio data to the temporary file
-        # Different models have different output formats, we'll try common keys
-        if 'audio' in speech:
-            # Convert numpy array to WAV file
-            try:
-                import scipy.io.wavfile as wavfile
-                wavfile.write(temp_filename, speech['sampling_rate'], speech['audio'])
-            except ImportError:
-                # If scipy is not available, try raw writing
-                with open(temp_filename, 'wb') as f:
-                    # Convert numpy array to bytes in a simple way
-                    if isinstance(speech['audio'], np.ndarray):
-                        audio_bytes = speech['audio'].tobytes()
-                        f.write(audio_bytes)
-                    else:
-                        f.write(speech['audio'])
-        elif 'numpy_array' in speech:
-            with open(temp_filename, 'wb') as f:
-                f.write(speech['numpy_array'].tobytes())
-        else:
-            # Fallback: try to write whatever is available
-            with open(temp_filename, 'wb') as f:
-                # Just write the first value that seems like it could be audio data
-                for key, value in speech.items():
-                    if isinstance(value, (bytes, bytearray)) or (
-                            isinstance(value, np.ndarray) and value.size > 1000):
-                        if isinstance(value, np.ndarray):
-                            f.write(value.tobytes())
-                        else:
-                            f.write(value)
-                        break
+        temp_file.close()  # Close the file so we can write to it
         
+        # Write the raw audio data to the file
+        with open(temp_filename, 'wb') as f:
+            f.write(speech['bytes'])  # Using the 'bytes' field instead of 'audio'
+            
         return temp_filename
         
     except Exception as e:
         st.error(f"Error generating audio: {str(e)}")
-        # Print all available keys for debugging
         return None
 
-# Let's try a simpler approach with a functioning TTS model
-def simple_text2audio(story_text):
-    """Simplified version that just returns a hardcoded audio file"""
-    # In a real application, you'd use a working TTS model
-    # For demonstration, we'll create a simple audio file with a message
-    
-    # Create a placeholder WAV file (just 1 second of silence)
-    temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.wav')
-    temp_filename = temp_file.name
-    temp_file.close()
-    
-    # Generate a very simple silent WAV file
-    with open(temp_filename, 'wb') as f:
-        # Simple WAV header for 1 second of silence at 16000Hz
-        # RIFF header
-        f.write(b'RIFF')
-        f.write((36).to_bytes(4, byteorder='little'))  # File size - 8
-        f.write(b'WAVE')
-        
-        # Format chunk
-        f.write(b'fmt ')
-        f.write((16).to_bytes(4, byteorder='little'))  # Chunk size
-        f.write((1).to_bytes(2, byteorder='little'))   # PCM format
-        f.write((1).to_bytes(2, byteorder='little'))   # Mono
-        f.write((16000).to_bytes(4, byteorder='little'))  # Sample rate
-        f.write((32000).to_bytes(4, byteorder='little'))  # Byte rate
-        f.write((2).to_bytes(2, byteorder='little'))   # Block align
-        f.write((16).to_bytes(2, byteorder='little'))  # Bits per sample
-        
-        # Data chunk
-        f.write(b'data')
-        f.write((32000).to_bytes(4, byteorder='little'))  # Data size
-        
-        # 1 second of silence (16000 samples at 16-bit)
-        silence = bytes(32000)
-        f.write(silence)
-    
-    return temp_filename
-
 # Function to save temporary image file
 def save_uploaded_image(uploaded_file):
     if not os.path.exists("temp"):
@@ -192,10 +111,7 @@ if uploaded_file is not None:
     
     # Stage 3: Story to Audio data
     st.text('Generating audio data...')
-    # Uncomment the next line to try the complex implementation
-    # audio_file = text2audio(story)
-    # Use the simple implementation for now
-    audio_file = simple_text2audio(story)
+    audio_file = text2audio(story)
     
     # Play button
     if st.button("Play Audio"):
@@ -208,6 +124,6 @@ if uploaded_file is not None:
     # Clean up the temporary files
     try:
         os.remove(image_path)
-        # We'll leave the audio file for playback
+        # Don't delete audio file immediately as it might still be playing
     except:
         pass